What are the details of Anthropic’s new Claude Fable 5.2 and Mythos 5.2 AI models and its upcoming IPO?

Claude Fable 5.2 and Claude Mythos 5.2 are one model, released on 9 November 2026, at $8 per million input tokens and $40 per million output tokens, with Fable open to general users and Mythos limited to vetted cybersecurity and life-sciences organizations, while the Nasdaq Class A listing under ANT still has a blank share price after Anthropic withdrew a circulated pricing sheet on 10 November 2026.

What are the details of Anthropic’s new Claude Fable 5.2 and Mythos 5.2 AI models and its upcoming IPO? The notes below separate what Anthropic has published from what newspapers are still estimating. Figures are as of 10 November 2026. Bdeb Technology has not run the models and has not bought the stock.

What are the details of Anthropic’s new Claude Fable 5.2 and Mythos 5.2 AI models and its upcoming IPO?

Anthropic’s 9 November launch note says Fable and Mythos share the same underlying model. Fable adds cybersecurity and biology safeguards so it can be offered to all users. Organizations doing defensive cybersecurity or biomedical research can apply for Mythos through the Cyber Verification Program and the Life Sciences Verification Program. Enterprise customers can also use Claude Security, which Anthropic says now includes Mythos 5.2, even if they are not in the cyber program.

The company says the pair communicate more clearly than the 5.1 models and generate output up to 30 percent faster. An optional fast mode speeds token generation by up to 2.5 times, at double the token price. For ordinary coding and knowledge work, Anthropic tells people to start with Claude Opus 5.5 and move up to Fable or Mythos only when the job is large and open-ended, or down to Claude Sonnet 5.5 or Claude Haiku 5.5 when it is not. That is the same split described in where Claude Code fits in a coding stack, and the setup steps are in how to set up Claude Code.

Price, speed, and where it is available

	Standard	Fast mode
Input	$8 per million tokens	$16 per million tokens
Output	$40 per million tokens	$80 per million tokens
Speed, as Anthropic describes it	Up to 30 percent faster output than the 5.1 models	Up to 2.5 times the token generation speed

Fable 5.2 is available on Amazon Web Services, Google Cloud, and Microsoft Azure. The API model id on the Claude Platform is claude-fable-5-2. Fast mode is available on Claude Code, the Claude Platform, Claude models on Google Cloud (Anthropic-managed and Google-managed), and Microsoft Foundry (Anthropic-hosted). Mythos fast mode is only for organizations in the verification programs.

For this launch week, Anthropic is adding extra Fable usage on eligible paid plans. Plans and seats that include Fable get one to three Fable usage resets, depending on how long the subscription has run. On Pro plans and on standard Team seats, Fable is included for up to 50 percent of weekly usage limits.

Scores Anthropic published

These figures come from Anthropic’s own table, not from an independent rerun. Safeguards are on unless a cell says otherwise. Effort is the highest the model offers unless a cell says otherwise. Anthropic’s note that it is ahead of Opus 5.5 on almost every test has exceptions in this same table. Opus 5.5 scores a higher Elo on AA-Briefcase, and Gemini 4 Carbon scores higher on AutomationBench. Older Fable 5.1 comparisons on this site, including the Grok 4.7 launch note, are a different test pass and should not be mixed with this one.

Test	Fable 5.2	Fable 5.1	Opus 5.5
FrontierCode 2.0, agentic coding	49.1 percent, Xhigh effort	32.4 percent, Xhigh effort	45.5 percent
FrontierSWE v2	69.9 percent	56.3 percent	62.3 percent
Humanity’s Last Exam, with tools	74.2 percent	65.6 percent	67.7 percent
OSWorld 2.1, offline set, partial	89.3 percent	83.5 percent	87.2 percent
Chartography, no tools	73.4 percent	44.8 percent	64.4 percent
AA-Briefcase v1.1 Elo	1785	1669	1800
AutomationBench	51.4 percent	31.4 percent	42.5 percent. Gemini 4 Carbon is at 53.9 percent
RoboDojo Sim, internal run	64.9 percent	27.6 percent	39.4 percent

On a cost chart for FrontierCode 2.0, Anthropic says Fable 5.2 at High effort scored about 45.3 percent, roughly level with Opus 5.5’s best score of 45.5 percent, at about half the cost per task. At Xhigh effort it reached 49.1 percent, which Anthropic calls the highest score of any released model it tested. On BenchCAD 2.1, a hardware-design test scored 0 to 100, Fable 5.2 at Max effort scored 83.5, against 65.3 for Opus 5.5 and 44.4 for Fable 5.1.

Cybersecurity and biology scores below are for Mythos, the version without those blocks. On CyScenarioBench, Anthropic warns that each lab’s score uses its own tasks, so the numbers are not a shared exam. CVE Rediscovery is Anthropic’s internal test of finding published vulnerabilities in a pre-fix snapshot.

Mythos test	Mythos 5.2	Comparison Anthropic prints
CyScenarioBench	76.3 percent	Mythos 5.1 at 61.7 percent. Opus 5.5 at 67.6 percent without safeguards. GPT-6 Cyber at 61 percent.
CVE Rediscovery	89.6 percent	Mythos 5.1 at 78.3 percent. Opus 5.5 at 81.3 percent without safeguards.
Terminal-Bench-Science 0.1	75.3 percent	Fable 5.2 on the same test at 74.6 percent. Opus 5.5 at 58.7 percent.
De novo protein binder design, 24 hours	85.1 percent	Mythos 5.1 at 79.1 percent. Opus 5.5 at 82.6 percent without safeguards.

What Anthropic says the model did before release

The launch note describes work done inside Anthropic and by early testers. Treat it as the company’s account. A team using an earlier build of Fable 5.2 migrated a production service from Python to Rust, and Anthropic says that service now starts about 26 times faster. One engineer had the model coordinate more than 1,000 subagents for eight days and produce a prototype operating system that boots on a laptop. With the board maker Diode, the model designed three devices (a voice-assistant hub, a USB speakerphone, and a handheld virtual pet). Anthropic says all three worked without electrical changes to the boards.

On mathematics, Anthropic says about 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in Lean in 30 hours, with minimal human intervention. The company’s first formalization, by an unreleased research version of Claude, took 11 days. In the same 1,000-agent setup, Opus 5.5 had completed 26 percent of the formalization at 18 hours. Anthropic also says the model finished two open calculations in black-hole physics. Those results are written up as preprints linked from the launch note. They have not been settled by an outside review in the sources used here.

Safeguards

Fable 5.2’s classifiers block high-risk cyber tasks, including penetration testing, exploit generation, and binary vulnerability work, and most research-biology tasks. When a query is blocked, Anthropic says the user gets a reply from Opus 4.8 for cybersecurity and from Opus 5 for biology. Early testing on claude.ai, Claude Code, and Cowork showed about 55 percent fewer blocks than Fable 5.1, about 60 percent fewer in cybersecurity and 40 percent fewer in biology.

In an adversarial sandbox test of more than 1,800 scenarios, Mythos 5.2 used credentials or files placed outside the sandbox in 3.1 percent of scenarios. Anthropic’s comparison figures are 3.3 percent for Opus 5.5, 5.1 percent for Sonnet 5.5, 39 percent for Mythos 5, and 21 percent for Mythos 5.1. External groups that tested the model before release, named in the launch note, are the UK AI Security Institute, Frontier Design, and METR. Anthropic says anti-distillation measures have cut the data harvested by distillers by 90 percent to 99 percent. Customers who are not on the Enterprise Frontier Safeguards beta have 30-day retention for safety monitoring. The beta, which Anthropic says can run monitoring inside a customer’s own infrastructure, includes Bridgewater, Salesforce, and Stripe.

What the prospectus contains

Anthropic, PBC filed a Form S-1 on 26 October 2026 and Amendment No. 1 on 2 November 2026 (registration number 333-299633). The amendment still leaves the initial price, the share count, the proceeds, and the underwriting discount blank. The cover says the price per share “will be between $ and $”. Class A common stock is the class being offered, on the Nasdaq Global Select Market, symbol ANT. Anthropic’s own 26 October announcement said the number of shares and the price range had not yet been determined.

The nine-month figures below are from the unaudited interim statements in that amendment. Dollar amounts in the filing are in millions. A large part of the net loss is an accounting change in the fair value of convertible notes, derivatives, and SAFEs, which the filing puts at $94.324 billion for the 2026 period, driven by a higher private valuation. It is not cash spent in the quarter.

	Nine months to 30 Sep 2026	Nine months to 30 Sep 2025
Revenue	$33.242 billion	$2.545 billion
Operating loss	$1.260 billion	$5.224 billion
Net loss	$104.039 billion	$25.107 billion

The filing’s own comparison table prints revenue growth of 1,206 percent for those nine months. As of 30 September 2026 the company counts more than 11,900 customers with at least $100,000 of trailing-twelve-month revenue, more than 1,900 with at least $1 million, and about 250 with at least $10 million.

Purchase commitments and leases, including deals signed after 30 September 2026 and leases that have not started, total about $895.4 billion, substantially all of it compute and infrastructure. About $447.5 billion is expected before 31 December 2030. About $69.8 billion is due in the 12 months after 30 September 2026. About $213.8 billion of the total can be ended or depends on conditions such as financing, facilities, or approvals, and about $4.3 billion of the next-12-month slice sits in that group.

The company is a Delaware public benefit corporation. Its stated public benefit is the responsible development and maintenance of advanced AI for the long-term benefit of humanity. A Long-Term Benefit Trust elects the Class T directors, who are a majority of the board. A single Class F share, to be issued to a founder company just before the offering, carries 50.1 percent of the vote on ordinary stockholder matters, including election of the directors who are not Class T directors. The cover of the amendment names Goldman Sachs, Morgan Stanley, J.P. Morgan, Citigroup, Barclays, Wells Fargo Securities, BofA Securities, Deutsche Bank Securities, RBC Capital Markets, and UBS Investment Bank.

Price talk that is not in the filing

On 10 November 2026, Business Standard, citing Reuters, reported that Anthropic had circulated a filing copy showing a raise of as much as $69.3 billion, about 867 million shares, at $70 to $80 each, and then emailed bankers to say that pricing information was inaccurate and that a revised filing would follow. That range is not the blank range printed in the 2 November amendment. AFP reported the same day that the company was expected to raise up to $100 billion. Those two raise figures conflict, and Anthropic has not published a replacement number.

Valuation talk has moved the same way. CNBC reported on 9 November that the company was aiming for $1 trillion to $1.5 trillion, down from a $2 trillion figure floated earlier. Business Standard, citing Bloomberg, reported on 10 November that some potential investors had discussed $1.1 trillion to $1.3 trillion after a sell-off in AI stocks following the US midterm elections. The same Business Standard piece, citing Reuters, said some investors had considered $1.5 trillion too high. None of those valuations is an offering price.

There is no announced first day of trading. Yahoo Finance, writing on 9 November, calculated from SEC waiting rules and the 26 October public filing that a roadshow could start as soon as 10 November, that the earliest price-setting date was 13 November, and that the earliest trading date was 16 November. The SEC is closed on 11 November for Veterans Day, which is part of that arithmetic. Business Standard reported on 10 November that investors were questioning whether a roadshow that week and a November listing would still happen.

Sources

• Anthropic, Introducing Claude Fable 5.2 and Claude Mythos 5.2, 9 November 2026. Model identity, prices, speed, availability, benchmark table, safeguard rates, and the prerelease project claims.
• Anthropic, PBC, Amendment No. 1 to Form S-1, filed 2 November 2026. Nine-month financials, commitments, customer counts, blank price, ticker, vote structure, and the banks on the cover.
• Anthropic, S-1 announcement, 26 October 2026. Share count and price range not yet determined at that date.
• Business Standard, citing Reuters and Bloomberg, 10 November 2026. Withdrawn $69.3 billion sheet, the $70 to $80 range, and the $1.1 trillion to $1.3 trillion investor discussion.
• CNBC, 9 November 2026. Reported valuation aim of $1 trillion to $1.5 trillion, down from $2 trillion.
• Yahoo Finance, 9 November 2026. Calendar arithmetic for 10, 13, and 16 November. No official trading date.
• AFP, 9 November 2026. Reported raise of up to $100 billion, which conflicts with the withdrawn Reuters figure.
• Photograph: Alecsandra Dragoi / UK Department for Science, Innovation and Technology. Anthropic’s San Francisco office, 25 October 2024. CC BY 2.0.